Frequent item set mining method based on localized differential privacy

Abstract

The invention discloses a frequent item set mining method based on localized differential privacy, which comprises the following steps: 1, a local data protection stage: generating a disturbance parameter array according to the conditions of 0 and 1 on each bit in a 0 / 1 string after coding by adopting a self-adaptive coding strategy, and disturbing data by applying an immediate response technology according to the disturbance array; 2, a joint probability estimation stage: learning model parameters of the disturbed data through a hidden Markov model, and calculating and estimating a joint probability by using the parameters; and 3, a frequent item set discovery stage: constructing a probability dependency graph corresponding to the original data according to a result of the step 2, and searching a frequent item set in the probability graph through a frequent item set prior principle. According to the method, frequent item set mining considering data privacy of each user in a multi-user-side distribution environment can be supported.

Description

technical field [0001] The invention relates to a privacy protection data mining method, in particular to a frequent item set mining method based on localized differential privacy. Background technique [0002] In recent years, privacy-preserving frequent itemset mining has become a hot topic for researchers. Differential privacy has become a hot technology in the field of privacy protection because of its strict mathematical definition of protection effect and the need not to care about the background knowledge of the attacker. The traditional centralized differential privacy technology centralizes data to a third-party data center, but it is often difficult to find a trusted third-party data center in practical applications, so localized differential privacy technology is usually used. The distributed application scenario of the third-party data center protects the user's private data by perturbing the user's sensitive data at the user end. defined as follows [0003] D...

AI Technical Summary

Problems solved by technology

The existing localized differential privacy-preserving frequent itemset mining methods mainly have the following problems: (1) The traditional sampling-based frequent itemset mining method only samples one value at a time, and needs to interact with the client multiple times to obtain relevant data features, communication The cost is high, leading to an increase in the privacy budget and reducing the intensity of privacy protection; (2) The signal-to-noise ratio of the localized differential privacy data perturbation protocol is relatively large, and multi-dimensional data will generate a variety of different data combinations during data interaction. Every attribute will be perturbed, and the combination will undoubtedly amplify the noise in the perturbed data, reducing data availability; (3) The correlation of attributes in multidimensional data cannot be ignored, and the existing localized differential privacy frequent item mining technology is directly applied to Multidimensional data can lead to lower quality mining results

Images